 |
Center for Nonlinear Studies |
An increasingly promising technique for answering quantitative questions about the real world is the use of human volunteer work. These people can be experts or unskilled, and the work can be explicit (i.e., volunteers work on the problem directly) and/or implicit (volunteer do apparently unrelated tasks, and their work is transformed into the problem space). Achieving success requires addressing human-focused issues like incentives, noise related to myriad human imperfections, and various types of cost; essentially, one must go beyond a framing of relatively passive “citizen sensors” to richer models labeled “citizen science”, “human computation”, and others. This relatively informal talk is designed for researchers in any field and will have two key parts. First, I will give an overview of Cyclopath, a web map serving the bicycling community of the Minneapolis-St. Paul metro area. This system, which I co-founded and led the development of, is a “geowiki”: any user can edit both the base map as well as points of interest and other annotations, and changes are live immediately (i.e., they are peer-reviewed after publication, not before). Among other research results that I will touch on, user work within the system resulted in a 7% savings in the distance of computed bicycle routes, and combining fine-grained subjective preferences with objective user work (i.e., editing the bike map) can produce better routes than previous routing algorithms. Second, I will give a tour of other systems in this space; applications include bird observations (eBird),image labeling (ESP Game), and protein folding (FoldIt). Motivated by these examples, I’ll close with a discussion of my ongoing and potential work here at LANL to leverage human knowledge, skill, and behavior to make progress on our real-world global challenges.
Host: Kipton Barros, T-4 and CNLS